Skip to main content

Future Work

Future Work​

New Functionalities:​

  1. Motion Input Camera Integration
  2. Philips Hue light integration in the shader mode
  3. More diverse gesture detection triggering different effects
  4. Smoother shader mode particle reactions
  5. Background and texture generation applied in the shader mode
  6. Philips Hue dynamic light effect improvement with multiple colours across lights
  7. Extend the compatibility on other OS
  8. Convert IPC handlers in main file into separate IPC files for modularity

Existing Bugs:​

  1. Progress bar not working correctly sometimes in the shader mode
  2. Mouse intersection can be incorrect sometime

Optimisation of AI​

Some of the current generated AI images are slightly inappropriate for the children, although this was a decision process due to other models taking too long to generate on a CPU, a future optimisation may be to incorporate LORA so that the model can tailor it’s output to a specific. Furthermore, other models

Currently, the LLM takes a large time to generate as it is a 9B parameter model (even though it has been quantised), if supplied with a vector database of analysis points used to characterise the songs, it may provide better results. Furthermore, we did not have enough time to fine tune the model using model output from larger parameter models (distil tuning). This may allow for even smaller models to outperform the context analysis of the song itself rather than the lyrics, as the LLM current finds it hard to analyse the context of the song, making it generate somewhat out-of-tune theme colours. Allowing for a multi-modal approach with sample images of the song may allow for a more related set of chosen colours.